|
Abstract:
|
Hospital performance data are used for either reporting or incentive to improve health care quality. To allow for appropriate comparisons, hospitals are suggested to be compared within peer groups with similar combination of characteristics. Full peer group assignment, determined by clustering technique, are performed less frequently to make groups stable. For new hospitals to the comparison list, their groups are assigned by classification methods with existing hospitals as the training data set. In this study, we use four characteristic used for full assignment to predict three group assignments for new comers using R. A variety of classification methods are compared including both discriminant and many data mining techniques. Leave-one-out cross validation is applied for honest comparison of error rates among different methods. SVM, random forest, and bagging are three best methods.
|
